Three dimensions, not a secret type of person

In 2002 Delroy Paulhus and Kevin Williams did something obvious that hardly anyone had done before: they gave the same sample three questionnaires drawn from three separate research traditions. Narcissism came from clinical psychology, Machiavellianism from the social psychology of the 1960s, psychopathy from forensic research. Each tradition had its own literature, its own scales, and the quiet assumption that its construct was a thing of its own.

The sample consisted of students, not offenders or patients. The result fitted into half a sentence: “overlapping, but distinct”. Out of that half-sentence came the Dark Triad, and out of the Dark Triad came one of the most cited constructions in personality psychology. The argument that continues to this day concerns the second half of the sentence.

What the three scales are meant to separate

The emphases are set out concisely in the review by Adrian Furnham, Steven Richards and Paulhus. Subclinical narcissism stresses entitlement, status and self-enhancement. Machiavellianism means strategic, cynical, manipulative conduct with a long time horizon. Subclinical psychopathy covers emotional coldness, low remorse and often impulsivity. As the shared core, the authors name a cold instrumentalisation of others — “callous manipulation”.

“Subclinical” is not decoration here but the decisive qualification. What is measured is variation in general samples, not mental disorder. An elevated questionnaire score is not a diagnosis and proves neither abuse nor crime. People sit somewhere on scales, and behaviour additionally depends on opportunity, norms and consequences.

Where the three traits attach within the large personality models is not in dispute. Peter Muris and colleagues record in their critical review that all three bind primarily to a single dimension: to low agreeableness in the five-factor model, and to low honesty-humility in the HEXACO model. That finding has two faces. It shows that the triad captures something real — and at the same time that it already has a place in an established model.

How far the scales overlap

The average associations between the three traits fall between .34 and .58 according to Muris and colleagues. They are highest between Machiavellianism and psychopathy. The authors trace this not only to the subject matter but to the instruments. The Mach-IV, the SRP-III and the short-form SD3 have been criticised for failing to capture Machiavellianism and psychopathy as separate constructs at all — which inflates the correlations unduly.

The SD3 shows how much convenience matters here. Daniel Jones and Paulhus introduced it in 2014 as a short measure: 27 statements for all three traits together, against dozens of statements per trait in the lengthy predecessors. Four studies with 1,063 people in total, drawn from community and student samples, examined its structure, reliability and validity; the last checked the scores against ratings by acquaintances rather than relying on self-report alone. The instrument is therefore better tested than its reputation suggests. But a shorter scale changes nothing if two of the three constructs already overlap substantively — it merely makes the overlap cheaper to reproduce.

A meta-analysis by Colin Vize, Donald Lynam, Katherine Collison and Joshua Miller examined this systematically in 2018. It compared not only the intercorrelations (156 data sets) but the entire networks of relations between the three traits and other variables (159 data sets). The result: the networks of psychopathy and Machiavellianism largely coincide, while narcissism shows relations that diverge. The finding held even when the data were separated by the instrument used. The authors draw an unusually direct conclusion: the existing Machiavellianism literature is better understood as a literature on secondary psychopathy.

Miller and colleagues gathered the methodological objections a year later. There are five. First, the three constructs are usually treated as one-dimensional although they are demonstrably multidimensional. Second, the common measures do not separate Machiavellianism and psychopathy. Third, the usual multivariate analytic procedures create statistical and interpretive difficulties. Fourth, the three traits are rarely tested directly against one another. Fifth, the field rests heavily on convenience samples and on a single method of data collection — self-report.

The last point is the least conspicuous and the most consequential. Whoever sets out to measure whether someone coldly uses other people generally asks that person.

One dark factor instead of three traits

If the scales overlap this strongly, a more radical question follows: are they measuring the same thing at bottom? Morten Moshagen, Benjamin Hilbig and Ingo Zettler proposed exactly that in 2018. Their Dark Factor of Personality, D for short, is conceived as a general dispositional tendency of which the individual dark traits are only particular expressions.

Their definition is remarkably precise: the tendency to maximise one’s own utility while accepting, tolerating or maliciously bringing about disadvantages for others — accompanied by beliefs that justify it. The second part carries the main weight. What constitutes the factor is not coldness alone but the supplied rationale for why the harm is all right. Across four studies with more than 2,500 people in total, the authors examined a larger number of dark traits simultaneously and found that these can be understood as specific manifestations of D. Which individual scale was used made no decisive difference to the result.

The list is no more closed at the upper end. Erin Buckels, Daniel Jones and Paulhus tested in 2013 whether everyday sadism belongs alongside the rest. In one study, participants could choose to kill insects; people scoring high on sadism did so more often. A second study concerned harming an uninvolved person. Where aggression was easy to come by, both sadism and the triad measures predicted it. But only the sadists were willing to do extra work for the opportunity to cause harm. The triad became a tetrad — and the argument about how many dark traits there are shows above all that the number three was never the finding.

Patterns rather than monsters

Online, all of this regularly turns into a machine for revelations: the ex is a narcissist, the colleague Machiavellian, the boss a psychopath. In retrospect every hurtful relationship supplies fitting examples. Without a comparison group, informant reports, a time course and the question of actual impairment, the analysis becomes a Barnum effect with dark vocabulary — a horoscope whose sentences land only because they fit almost anyone.

For decisions about protection, observable patterns therefore matter more than labels, and repeated ones at that. Are boundaries violated more than once? Are mistakes acknowledged? Do words and actions match over time? Is closeness used against the person? Is there repair after harm has been done? These questions need no diagnosis. Counterevidence matters just as much: genuine amends, stable care, and the ability to limit one’s own advantage for someone else’s sake. No single observation reliably separates the three paths — research cannot manage it with hundreds of data sets, and a dinner certainly cannot.

What remains

The shared core is secure. There is an antagonistic underlying tendency, it is measurable, and it predicts problematic behaviour. It sits where the established models locate low agreeableness and low honesty-humility respectively. What is contested is the partition. Whether there are three traits, four, or a single factor with several faces has not been settled after two decades. Part of that dispute goes back to the questionnaires, not to the people who fill them in.

In everyday life you can spot the misuse by one simple sign: it describes a person rather than an action, and it does so in hindsight. The triad then no longer explains why someone did something; it rules out that they could have acted otherwise. Personality traits raise probabilities. They do not write a script — and they release nobody from judging specific actions specifically.